The initial search yielded 3,901 non-duplicated studies. A total of 110 studies were identified as potentially meeting our inclusion criteria and full-text of the studies were reviewed. After reviewing, another 79 studies were further excluded mainly due to lack of extractable or calculable incidence data, or different study population (ie. none general population-based study). Finally, a total of 31 studies were included in this meta-analysis (Figure 1). List of included studies were showed in supplementary file 2.
The characteristics of individual study were summarized in table 1 and supplementary table 1. The quality assessment of each paper included in the study was shown in supplementary table 2. The average score in the quality assessment was 4.8. The major risks of bias existed in “blinding to demographic variables” domain due to blind the person/panel who diagnosed cases to certain demographic characteristics, and “leakage study” domain due to lack of attempting to identify the original cases.
The pooled annual incidence of DILI
The overall incidence of DILI among the general population was 4.94 (95% CI: 4.05-5.83) per 100,000 person-years (14 studies,supplementary file 2 for reference list). There was high heterogeneity among the studies (I2=99%, P<0.01, Supplementary Figure 1). Figure 2 showed the incidence of DILI of each country with at least one study.
Pooled result of five studies showed the incidence of DILI in terms of hepatocellular, cholestatic and mixed patterns was 1.51 (95%CI: 0.72-2.29), 0.78 (95%CI: 0.37-1.18) and 0.58 (95%CI: 0.21-0.96) per 100,000 person-years, respectively (Supplementary Figure 2).
The DILI incidence varied by regions, as noted in table 2. The incidence was 17.82 (95%CI: 6.26-29.38) per 100,000 person-years (2 studies) in Asia, 3.89 (95%CI: 2.81-4.97) per 100,000 person-years in Europe (7 studies), and 1.72 (95%CI: 0.48-2.95) per 100,000 person-years in America (5 study).
Time-based cumulative meta-analysis showed an overall increase of the incidence of DILI over time (Figure 3). Subgroup analysis showed that the highest incidence among studies whose patients’ enrollment were after 2010 (15.14 per 100,000 person-years, 95%CI: 0.00-30.39) compared to those enrolling patients at 2000-2010 (4.44 per 100,000 person-years, 95%CI: 0.84-8.05), and before 2000 (2.55 per 100,000 person-years, 95%CI: 1.63-3.47).
We found that the DILI incidence varied among the reporting sources. The incidence was 11.78 per 100,000 person-years (95%CI: 1.32-22.24) among studies whose DILI cases were reported by physicians and 9.57 per 100,000 person-years (95%CI: 5.06-14.09) among those whose cases were identified based on hospital case series, compared to 1.61 per 100,000 person-years (95%CI: 0.82-2.41) for studies using insurance/administrative databases. Moreover, the pooled estimates of DILI incidence were higher in studies with prospective design and studies with quality score<5.
Incidence of DILI: Subgroup analysis by age and gender
The DILI incidence by age group was not pooled in this meta-analysis because of the high variability of age groups used. Nonetheless, consistent higher incidence in old age group was observed (Supplementary Figure 3). For example, in the study conducted in Iceland, the incidence of DILI ranged from 8.5 per 100,000 person-years (95% CI: 3.7-16.8) in people aged 15-24 years to 41 per 100,000 person-years (95% CI: 18.7-77.8) in people aged 80-106 years.16
Five studies reported the incidence of DILI in male and female separately, but with inconsistent results (Supplementary Figure 3). The pooled results of these studies showed similar incidence of DILI between male (3.42 per 100,000 person-years, 95%CI: 1.83-5.00) and female (4.64 per 100,000 person-years, 95%CI: 2.74-6.55). Heterogeneity was substantial in both female and male sets (I2=96%, I2=98%).
Incidence of DILI: By given drug(s)
As to the incidence of drug-specific DILI among the users, we found that the incidence of statins-induced DILI was 11.30 per 100,000 person-years (95% CI: 6.48-19.69, five studies). Notably, rosuvastatin users had a slighter lower incidence of DILI (8.31 per 100,000 person-years, 95% CI: 2.21-31.18) compared with other statins users (12.60 per 100,000 person-years, 95% CI: 7.44-21.34) (Table 3) (supplementary file 2 for reference list).
The incidence of antifungal, antidepressants, paracetamol and antidiabetic drugs induced DILI was 63.46 (95%CI: 33.39-120.57, three studies), 3.83 (95%CI: 0.97-15.03, three studies), 0.16 (95%CI: 0.02-1.08, two studies), and 18.20 (95%CI: 1.51-219.20, two studies) per 100,000 person-years, respectively. Only one study reported the incidence of anti-TB, non-steroidal anti-inflammatory drugs (NSAIDs), anti-thyroid drugs (ATD), and oral iron chelator induced liver injury, respectively.
Univariate and multivariate meta-regression analysis was conducted to explore the reasons accounting for the high heterogeneity observed in our meta-analysis (Supplementary Table 3). The results showed that region distribution could partly explain this high heterogeneity among the studies (P<0.05).
Sensitivity analysis and publication bias
The pooled estimate was 7.42 per 100,000 person-years (95%CI: 5.52-9.32) for studies with case ascertainment by causality assessment, 5.90 per 100,000 person-years (95%CI: 4.85-6.95) for studies published after the detection of hepatitis C of 1990 year, and 6.02 per 100,000 person-years (95%CI: 4.85-7.18) after removing the study of Goldberg DS et al. The sensitivity analysis showed a slightly increase in the pooled estimates, mainly because most of the studies included in the sensitivity analysis were relatively recent published studies. This was consistent with the temporal trend of DILI incidence over time.
There was no evidence of publication bias as assessed neither by Begg’s nor Egger’s test for meta-analyses (p=0.32 and p=0.08).